Reformat count operator (#6432)

This commit reformats the count operator so the record expression is no
longer required and the count alias in the record expression is now the
last argument.

Author: mattnibs

compiler/parser/parser.go

@@ -485,6 +485,13 @@ CountOp
       return &ast.CountOp{
         Kind: "CountOp",
         Expr: rec.(*ast.RecordExpr),
+        Loc: loc(c),
+      }, nil
+    }
+  / COUNT &EndOfOp {
+      return &ast.CountOp{
+        Kind: "CountOp",
+        Loc: loc(c),
       }, nil
     }
 

compiler/parser/parser.peg

@@ -88,12 +88,14 @@ func (c *checker) op(typ super.Type, op sem.Op) super.Type {
 	case *sem.BadOp:
 		return c.unknown
 	case *sem.CountOp:
-		elems := []sem.RecordElem{
-			&sem.FieldElem{Name: op.Alias, Value: &sem.LiteralExpr{Value: "0(uint64)"}},
-		}
+		var elems []sem.RecordElem
 		if op.Expr != nil {
 			elems = append(elems, op.Expr.(*sem.RecordExpr).Elems...)
 		}
+		elems = append(elems, &sem.FieldElem{
+			Name:  op.Alias,
+			Value: &sem.LiteralExpr{Value: "0(uint64)"},
+		})
 		return c.recordElems(typ, elems)
 	case *sem.CutOp:
 		return c.pathsToType(c.assignments(typ, op.Args))

compiler/semantic/checker.go

@@ -652,23 +652,38 @@ func (t *translator) semOp(o ast.Op, seq sem.Seq) sem.Seq {
 			Cases: cases,
 		})
 	case *ast.CountOp:
-		if len(o.Expr.Elems) == 0 {
-			t.error(o.Expr, errors.New("count record expression must not be empty"))
-			return append(seq, badOp())
-		}
-		first, ok := o.Expr.Elems[0].(*ast.ExprElem)
-		if !ok {
-			t.error(o.Expr, errors.New("first element in record expression for count must be an identifier"))
-			return append(seq, badOp())
-		}
-		alias := first.Expr.(*ast.IDExpr).Name
+		var alias string
 		var expr sem.Expr
-		if len(o.Expr.Elems) > 1 {
-			expr = t.expr(&ast.RecordExpr{
-				Kind:  "RecordExpr",
-				Elems: o.Expr.Elems[1:],
-				Loc:   o.Expr.Loc,
-			})
+		if o.Expr == nil {
+			alias = "count"
+			expr = &sem.RecordExpr{
+				Elems: []sem.RecordElem{
+					&sem.FieldElem{Name: "that", Value: sem.NewThis(nil, nil)},
+				},
+			}
+		} else {
+			n := len(o.Expr.Elems)
+			if n == 0 {
+				t.error(o.Expr, errors.New("count record expression must not be empty"))
+				return append(seq, badOp())
+			}
+			last := o.Expr.Elems[n-1]
+			if exprElem, ok := last.(*ast.ExprElem); ok {
+				if id, ok := exprElem.Expr.(*ast.IDExpr); ok {
+					alias = id.Name
+				}
+			}
+			if alias == "" {
+				t.error(last, errors.New("last element in record expression for count must be an identifier"))
+				return append(seq, badOp())
+			}
+			if len(o.Expr.Elems) > 1 {
+				expr = t.expr(&ast.RecordExpr{
+					Kind:  "RecordExpr",
+					Elems: o.Expr.Elems[:n-1],
+					Loc:   o.Expr.Loc,
+				})
+			}
 		}
 		return append(seq, &sem.CountOp{
 			Node:  o,

compiler/semantic/op.go

@@ -411,12 +411,12 @@ func (c *canonDAG) op(p dag.Op) {
 		c.write("combine")
 	case *dag.CountOp:
 		c.next()
-		c.write("count {%s", p.Alias)
+		c.write("count {")
 		if p.Expr != nil {
-			c.write(",")
 			c.recordElems(p.Expr.(*dag.RecordExpr).Elems)
+			c.write(",")
 		}
-		c.write("}")
+		c.write("%s}", p.Alias)
 	case *dag.CutOp:
 		c.next()
 		c.write("cut ")

compiler/sfmt/dag.go

@@ -1,13 +1,13 @@
 # Test ensures that flowgraphs utilizing the count operator
 # are not parallelized.
 script: |
-  SUPER_VAM=1 super compile -C -P 2 'from /dev/null | count {row_number,...this} | aggregate sum(x)'
+  SUPER_VAM=1 super compile -C -P 2 'from /dev/null | count {...this,row_number} | aggregate sum(x)'
 
 outputs:
   - name: stdout
     data: |
       file /dev/null unordered
-      | count {row_number,...this}
+      | count {...this,row_number}
       | aggregate
           sum:=sum(x)
       | values sum

compiler/ztests/par-count.yaml

@@ -16,10 +16,10 @@ type Op struct {
 func New(sctx *super.Context, parent sbuf.Puller, alias string, in expr.Evaluator) (*Op, error) {
 	o := &Op{parent: parent, alias: alias}
 	var elems []expr.RecordElem
-	elems = append(elems, expr.RecordElem{Name: alias, Field: evalfunc(o.evalCount)})
 	if in != nil {
 		elems = append(elems, expr.RecordElem{Spread: in})
 	}
+	elems = append(elems, expr.RecordElem{Name: alias, Field: evalfunc(o.evalCount)})
 	var err error
 	o.expr, err = expr.NewRecordExpr(sctx, elems)
 	return o, err

runtime/sam/op/count/count.go

@@ -17,10 +17,10 @@ type Count struct {
 func NewCount(sctx *super.Context, parent vector.Puller, alias string, in expr.Evaluator) *Count {
 	o := &Count{parent: parent, alias: alias}
 	var elems []expr.RecordElem
-	elems = append(elems, expr.RecordElem{Name: alias, Expr: evalfunc(o.evalCount)})
 	if in != nil {
 		elems = append(elems, expr.RecordElem{Expr: in})
 	}
+	elems = append(elems, expr.RecordElem{Name: alias, Expr: evalfunc(o.evalCount)})
 	o.expr = expr.NewRecordExpr(sctx, elems)
 	return o
 }

runtime/vam/op/count.go

@@ -1,45 +1,56 @@
 script: |
+  super -s -c 'values {x:1},{x:2} | count'
+  echo // ===
   super -s -c 'values null,null | count {count}'
   echo // ===
-  super -s -c 'values {x:1},{x:2} | count {row_number,...this}'  
+  super -s -c 'values {x:1},{x:2} | count {...this,row_number}'
   echo // ===
-  super -s -c 'values {x:1,y:1},{x:2,y:2} | count {row_number,x}'  
+  super -s -c 'values {x:1,y:1},{x:2,y:2} | count {x,row_number}'  
   echo // ===
-  super -s -c 'values [1,2],[3,4] | unnest this into ( count {c,val:this} )'
-  ! super -s -c 'values null | count {count} | count := count+"hello"'
+  super -s -c 'values [1,2],[3,4] | unnest this into ( count {val:this,c} )'
+  ! super -s -c 'values null | count | count := count+"hello"'
   >&2 echo // ===
-  ! super -s -c 'values {x:"foo"} | count {row,...this} | x := x+1'
+  ! super -s -c 'values {x:"foo"} | count {...this,row} | x := x+1'
   >&2 echo // ===
   ! super -s -c 'values {x:"foo"} | count {...this}'
+  >&2 echo // ===
+  ! super -s -c 'count {foo:"foo",1}'
 
 vector: true
 
 outputs:
   - name: stdout
     data: |
+      {that:{x:1},count:1::uint64}
+      {that:{x:2},count:2::uint64}
+      // ===
       {count:1::uint64}
       {count:2::uint64}
       // ===
-      {row_number:1::uint64,x:1}
-      {row_number:2::uint64,x:2}
+      {x:1,row_number:1::uint64}
+      {x:2,row_number:2::uint64}
       // ===
-      {row_number:1::uint64,x:1}
-      {row_number:2::uint64,x:2}
+      {x:1,row_number:1::uint64}
+      {x:2,row_number:2::uint64}
       // ===
-      {c:1::uint64,val:1}
-      {c:2::uint64,val:2}
-      {c:1::uint64,val:3}
-      {c:2::uint64,val:4}
+      {val:1,c:1::uint64}
+      {val:2,c:2::uint64}
+      {val:3,c:1::uint64}
+      {val:4,c:2::uint64}
   - name: stderr
     data: |
-      type mismatch at line 1, column 40:
-      values null | count {count} | count := count+"hello"
-                                             ~~~~~~~~~~~~~
+      type mismatch at line 1, column 32:
+      values null | count | count := count+"hello"
+                                     ~~~~~~~~~~~~~
       // ===
       type mismatch at line 1, column 47:
-      values {x:"foo"} | count {row,...this} | x := x+1
+      values {x:"foo"} | count {...this,row} | x := x+1
                                                     ~~~
       // ===
-      first element in record expression for count must be an identifier at line 1, column 26:
+      last element in record expression for count must be an identifier at line 1, column 27:
       values {x:"foo"} | count {...this}
-                               ~~~~~~~~~
+                                ~~~~~~~
+      // ===
+      last element in record expression for count must be an identifier at line 1, column 18:
+      count {foo:"foo",1}
+                       ~